A parolee classification system using discriminant analysis is presented. Approximately 13,000 parolees released in 1969 and followed for two years constitute the data base. The sample is split into two parts so that the model can be first tested and then validated. The results of both the test and holdout samples suggest that the discriminant model is quite useful for classifying parolees into "good" versus "poor" parole risks.When we compare the multivariate results with those obtained using univariate analysis, we show that the multivariate results are superior to the univariate results and demonstrate that results generated from univariate analyses can be quite misleading. In particular, it is demonstrated that the "best" univariate discriminator is a relatively poor multivariate discriminator. Moreover, the best multivariate discriminator appears to be a far less important discriminator in a univariate framework.The use of multivariate techniques to explain differential parolee performance was advocated more than fifty years ago (Hart, 1923). Nevertheless, many researchers continue to utilize univariate analyses when examining parolee performance (Glaser, 1964; MacSpeiden, 1966; Pownall, 1966;Gottfredson et al., 1970).' Univariate techniques are unlikely to provide decision makers with as much information as multivariate techniques. The former methods, unlike the latter, do not account for the interrelationships among the various factors which influence parolee performance. Often, variables which appear important in a univariate context become insignificant in multivariate frameworks; similarly, variables appearing unimportant in univariate frameworks often are shown to be important in multivariate contexts.2 Hence, information derived from univariate analyses can be misleading, causing decision makers to err more than they would with properly applied multivariate techniques.